Agents ↗

Editorial Notes

This is the foundational doc for OpenAI’s Agents SDK and defines how agents are constructed from instructions, tools, guardrails, and model configuration. Compare this instruction-based agent definition with CrewAI’s role/goal/backstory approach and LangGraph’s node-based agent composition — each framework makes different tradeoffs between declarative simplicity and fine-grained control. Pay attention to how guardrails are attached at the agent level rather than globally, which gives you per-agent safety policies but requires you to remember to configure them on every agent. Start here before reading the handoffs or tracing docs, as everything in the SDK builds on the Agent class.